1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Tests Memory Protection Keys (see Documentation/core-api/protection-keys.rst)
4  *
5  * There are examples in here of:
6  *  * how to set protection keys on memory
7  *  * how to set/clear bits in pkey registers (the rights register)
8  *  * how to handle SEGV_PKUERR signals and extract pkey-relevant
9  *    information from the siginfo
10  *
11  * Things to add:
12  *	make sure KSM and KSM COW breaking works
13  *	prefault pages in at malloc, or not
14  *	protect MPX bounds tables with protection keys?
15  *	make sure VMA splitting/merging is working correctly
16  *	OOMs can destroy mm->mmap (see exit_mmap()), so make sure it is immune to pkeys
17  *	look for pkey "leaks" where it is still set on a VMA but "freed" back to the kernel
18  *	do a plain mprotect() to a mprotect_pkey() area and make sure the pkey sticks
19  *
20  * Compile like this:
21  *	gcc -mxsave      -o protection_keys    -O2 -g -std=gnu99 -pthread -Wall protection_keys.c -lrt -ldl -lm
22  *	gcc -mxsave -m32 -o protection_keys_32 -O2 -g -std=gnu99 -pthread -Wall protection_keys.c -lrt -ldl -lm
23  */
24 #define _GNU_SOURCE
25 #define __SANE_USERSPACE_TYPES__
26 #include <errno.h>
27 #include <linux/elf.h>
28 #include <linux/futex.h>
29 #include <time.h>
30 #include <sys/time.h>
31 #include <sys/syscall.h>
32 #include <string.h>
33 #include <stdio.h>
34 #include <stdint.h>
35 #include <stdbool.h>
36 #include <signal.h>
37 #include <assert.h>
38 #include <stdlib.h>
39 #include <ucontext.h>
40 #include <sys/mman.h>
41 #include <sys/types.h>
42 #include <sys/wait.h>
43 #include <sys/stat.h>
44 #include <fcntl.h>
45 #include <unistd.h>
46 #include <sys/ptrace.h>
47 #include <setjmp.h>
48 
49 #include "pkey-helpers.h"
50 
51 int iteration_nr = 1;
52 int test_nr;
53 
54 u64 shadow_pkey_reg;
55 int dprint_in_signal;
56 char dprint_in_signal_buffer[DPRINT_IN_SIGNAL_BUF_SIZE];
57 
58 void cat_into_file(char *str, char *file)
59 {
60 	int fd = open(file, O_RDWR);
61 	int ret;
62 
63 	dprintf2("%s(): writing '%s' to '%s'\n", __func__, str, file);
64 	/*
65 	 * these need to be raw because they are called under
66 	 * pkey_assert()
67 	 */
68 	if (fd < 0) {
69 		fprintf(stderr, "error opening '%s'\n", str);
70 		perror("error: ");
71 		exit(__LINE__);
72 	}
73 
74 	ret = write(fd, str, strlen(str));
75 	if (ret != strlen(str)) {
76 		perror("write to file failed");
77 		fprintf(stderr, "filename: '%s' str: '%s'\n", file, str);
78 		exit(__LINE__);
79 	}
80 	close(fd);
81 }
82 
83 #if CONTROL_TRACING > 0
84 static int warned_tracing;
85 int tracing_root_ok(void)
86 {
87 	if (geteuid() != 0) {
88 		if (!warned_tracing)
89 			fprintf(stderr, "WARNING: not run as root, "
90 					"can not do tracing control\n");
91 		warned_tracing = 1;
92 		return 0;
93 	}
94 	return 1;
95 }
96 #endif
97 
98 void tracing_on(void)
99 {
100 #if CONTROL_TRACING > 0
101 #define TRACEDIR "/sys/kernel/tracing"
102 	char pidstr[32];
103 
104 	if (!tracing_root_ok())
105 		return;
106 
107 	sprintf(pidstr, "%d", getpid());
108 	cat_into_file("0", TRACEDIR "/tracing_on");
109 	cat_into_file("\n", TRACEDIR "/trace");
110 	if (1) {
111 		cat_into_file("function_graph", TRACEDIR "/current_tracer");
112 		cat_into_file("1", TRACEDIR "/options/funcgraph-proc");
113 	} else {
114 		cat_into_file("nop", TRACEDIR "/current_tracer");
115 	}
116 	cat_into_file(pidstr, TRACEDIR "/set_ftrace_pid");
117 	cat_into_file("1", TRACEDIR "/tracing_on");
118 	dprintf1("enabled tracing\n");
119 #endif
120 }
121 
122 void tracing_off(void)
123 {
124 #if CONTROL_TRACING > 0
125 	if (!tracing_root_ok())
126 		return;
127 	cat_into_file("0", "/sys/kernel/tracing/tracing_on");
128 #endif
129 }
130 
131 void abort_hooks(void)
132 {
133 	fprintf(stderr, "running %s()...\n", __func__);
134 	tracing_off();
135 #ifdef SLEEP_ON_ABORT
136 	sleep(SLEEP_ON_ABORT);
137 #endif
138 }
139 
140 /*
141  * This attempts to have roughly a page of instructions followed by a few
142  * instructions that do a write, and another page of instructions.  That
143  * way, we are pretty sure that the write is in the second page of
144  * instructions and has at least a page of padding behind it.
145  *
146  * *That* lets us be sure to madvise() away the write instruction, which
147  * will then fault, which makes sure that the fault code handles
148  * execute-only memory properly.
149  */
150 #ifdef __powerpc64__
151 /* This way, both 4K and 64K alignment are maintained */
152 __attribute__((__aligned__(65536)))
153 #else
154 __attribute__((__aligned__(PAGE_SIZE)))
155 #endif
156 void lots_o_noops_around_write(int *write_to_me)
157 {
158 	dprintf3("running %s()\n", __func__);
159 	__page_o_noops();
160 	/* Assume this happens in the second page of instructions: */
161 	*write_to_me = __LINE__;
162 	/* pad out by another page: */
163 	__page_o_noops();
164 	dprintf3("%s() done\n", __func__);
165 }
166 
167 void dump_mem(void *dumpme, int len_bytes)
168 {
169 	char *c = (void *)dumpme;
170 	int i;
171 
172 	for (i = 0; i < len_bytes; i += sizeof(u64)) {
173 		u64 *ptr = (u64 *)(c + i);
174 		dprintf1("dump[%03d][@%p]: %016llx\n", i, ptr, *ptr);
175 	}
176 }
177 
178 static u32 hw_pkey_get(int pkey, unsigned long flags)
179 {
180 	u64 pkey_reg = __read_pkey_reg();
181 
182 	dprintf1("%s(pkey=%d, flags=%lx) = %x / %d\n",
183 			__func__, pkey, flags, 0, 0);
184 	dprintf2("%s() raw pkey_reg: %016llx\n", __func__, pkey_reg);
185 
186 	return (u32) get_pkey_bits(pkey_reg, pkey);
187 }
188 
189 static int hw_pkey_set(int pkey, unsigned long rights, unsigned long flags)
190 {
191 	u32 mask = (PKEY_DISABLE_ACCESS|PKEY_DISABLE_WRITE);
192 	u64 old_pkey_reg = __read_pkey_reg();
193 	u64 new_pkey_reg;
194 
195 	/* make sure that 'rights' only contains the bits we expect: */
196 	assert(!(rights & ~mask));
197 
198 	/* modify bits accordingly in old pkey_reg and assign it */
199 	new_pkey_reg = set_pkey_bits(old_pkey_reg, pkey, rights);
200 
201 	__write_pkey_reg(new_pkey_reg);
202 
203 	dprintf3("%s(pkey=%d, rights=%lx, flags=%lx) = %x"
204 		" pkey_reg now: %016llx old_pkey_reg: %016llx\n",
205 		__func__, pkey, rights, flags, 0, __read_pkey_reg(),
206 		old_pkey_reg);
207 	return 0;
208 }
209 
210 void pkey_disable_set(int pkey, int flags)
211 {
212 	unsigned long syscall_flags = 0;
213 	int ret;
214 	int pkey_rights;
215 	u64 orig_pkey_reg = read_pkey_reg();
216 
217 	dprintf1("START->%s(%d, 0x%x)\n", __func__,
218 		pkey, flags);
219 	pkey_assert(flags & (PKEY_DISABLE_ACCESS | PKEY_DISABLE_WRITE));
220 
221 	pkey_rights = hw_pkey_get(pkey, syscall_flags);
222 
223 	dprintf1("%s(%d) hw_pkey_get(%d): %x\n", __func__,
224 			pkey, pkey, pkey_rights);
225 
226 	pkey_assert(pkey_rights >= 0);
227 
228 	pkey_rights |= flags;
229 
230 	ret = hw_pkey_set(pkey, pkey_rights, syscall_flags);
231 	assert(!ret);
232 	/* pkey_reg and flags have the same format */
233 	shadow_pkey_reg = set_pkey_bits(shadow_pkey_reg, pkey, pkey_rights);
234 	dprintf1("%s(%d) shadow: 0x%016llx\n",
235 		__func__, pkey, shadow_pkey_reg);
236 
237 	pkey_assert(ret >= 0);
238 
239 	pkey_rights = hw_pkey_get(pkey, syscall_flags);
240 	dprintf1("%s(%d) hw_pkey_get(%d): %x\n", __func__,
241 			pkey, pkey, pkey_rights);
242 
243 	dprintf1("%s(%d) pkey_reg: 0x%016llx\n",
244 		__func__, pkey, read_pkey_reg());
245 	if (flags)
246 		pkey_assert(read_pkey_reg() >= orig_pkey_reg);
247 	dprintf1("END<---%s(%d, 0x%x)\n", __func__,
248 		pkey, flags);
249 }
250 
251 void pkey_disable_clear(int pkey, int flags)
252 {
253 	unsigned long syscall_flags = 0;
254 	int ret;
255 	int pkey_rights = hw_pkey_get(pkey, syscall_flags);
256 	u64 orig_pkey_reg = read_pkey_reg();
257 
258 	pkey_assert(flags & (PKEY_DISABLE_ACCESS | PKEY_DISABLE_WRITE));
259 
260 	dprintf1("%s(%d) hw_pkey_get(%d): %x\n", __func__,
261 			pkey, pkey, pkey_rights);
262 	pkey_assert(pkey_rights >= 0);
263 
264 	pkey_rights &= ~flags;
265 
266 	ret = hw_pkey_set(pkey, pkey_rights, 0);
267 	shadow_pkey_reg = set_pkey_bits(shadow_pkey_reg, pkey, pkey_rights);
268 	pkey_assert(ret >= 0);
269 
270 	pkey_rights = hw_pkey_get(pkey, syscall_flags);
271 	dprintf1("%s(%d) hw_pkey_get(%d): %x\n", __func__,
272 			pkey, pkey, pkey_rights);
273 
274 	dprintf1("%s(%d) pkey_reg: 0x%016llx\n", __func__,
275 			pkey, read_pkey_reg());
276 	if (flags)
277 		assert(read_pkey_reg() <= orig_pkey_reg);
278 }
279 
280 void pkey_write_allow(int pkey)
281 {
282 	pkey_disable_clear(pkey, PKEY_DISABLE_WRITE);
283 }
284 void pkey_write_deny(int pkey)
285 {
286 	pkey_disable_set(pkey, PKEY_DISABLE_WRITE);
287 }
288 void pkey_access_allow(int pkey)
289 {
290 	pkey_disable_clear(pkey, PKEY_DISABLE_ACCESS);
291 }
292 void pkey_access_deny(int pkey)
293 {
294 	pkey_disable_set(pkey, PKEY_DISABLE_ACCESS);
295 }
296 
297 /* Failed address bound checks: */
298 #ifndef SEGV_BNDERR
299 # define SEGV_BNDERR		3
300 #endif
301 
302 #ifndef SEGV_PKUERR
303 # define SEGV_PKUERR		4
304 #endif
305 
306 static char *si_code_str(int si_code)
307 {
308 	if (si_code == SEGV_MAPERR)
309 		return "SEGV_MAPERR";
310 	if (si_code == SEGV_ACCERR)
311 		return "SEGV_ACCERR";
312 	if (si_code == SEGV_BNDERR)
313 		return "SEGV_BNDERR";
314 	if (si_code == SEGV_PKUERR)
315 		return "SEGV_PKUERR";
316 	return "UNKNOWN";
317 }
318 
319 int pkey_faults;
320 int last_si_pkey = -1;
321 void signal_handler(int signum, siginfo_t *si, void *vucontext)
322 {
323 	ucontext_t *uctxt = vucontext;
324 	int trapno;
325 	unsigned long ip;
326 	char *fpregs;
327 #if defined(__i386__) || defined(__x86_64__) /* arch */
328 	u32 *pkey_reg_ptr;
329 	int pkey_reg_offset;
330 #endif /* arch */
331 	u64 siginfo_pkey;
332 	u32 *si_pkey_ptr;
333 
334 	dprint_in_signal = 1;
335 	dprintf1(">>>>===============SIGSEGV============================\n");
336 	dprintf1("%s()::%d, pkey_reg: 0x%016llx shadow: %016llx\n",
337 			__func__, __LINE__,
338 			__read_pkey_reg(), shadow_pkey_reg);
339 
340 	trapno = uctxt->uc_mcontext.gregs[REG_TRAPNO];
341 	ip = uctxt->uc_mcontext.gregs[REG_IP_IDX];
342 	fpregs = (char *) uctxt->uc_mcontext.fpregs;
343 
344 	dprintf2("%s() trapno: %d ip: 0x%016lx info->si_code: %s/%d\n",
345 			__func__, trapno, ip, si_code_str(si->si_code),
346 			si->si_code);
347 
348 #if defined(__i386__) || defined(__x86_64__) /* arch */
349 #ifdef __i386__
350 	/*
351 	 * 32-bit has some extra padding so that userspace can tell whether
352 	 * the XSTATE header is present in addition to the "legacy" FPU
353 	 * state.  We just assume that it is here.
354 	 */
355 	fpregs += 0x70;
356 #endif /* i386 */
357 	pkey_reg_offset = pkey_reg_xstate_offset();
358 	pkey_reg_ptr = (void *)(&fpregs[pkey_reg_offset]);
359 
360 	/*
361 	 * If we got a PKEY fault, we *HAVE* to have at least one bit set in
362 	 * here.
363 	 */
364 	dprintf1("pkey_reg_xstate_offset: %d\n", pkey_reg_xstate_offset());
365 	if (DEBUG_LEVEL > 4)
366 		dump_mem(pkey_reg_ptr - 128, 256);
367 	pkey_assert(*pkey_reg_ptr);
368 #endif /* arch */
369 
370 	dprintf1("siginfo: %p\n", si);
371 	dprintf1(" fpregs: %p\n", fpregs);
372 
373 	if ((si->si_code == SEGV_MAPERR) ||
374 	    (si->si_code == SEGV_ACCERR) ||
375 	    (si->si_code == SEGV_BNDERR)) {
376 		printf("non-PK si_code, exiting...\n");
377 		exit(4);
378 	}
379 
380 	si_pkey_ptr = siginfo_get_pkey_ptr(si);
381 	dprintf1("si_pkey_ptr: %p\n", si_pkey_ptr);
382 	dump_mem((u8 *)si_pkey_ptr - 8, 24);
383 	siginfo_pkey = *si_pkey_ptr;
384 	pkey_assert(siginfo_pkey < NR_PKEYS);
385 	last_si_pkey = siginfo_pkey;
386 
387 	/*
388 	 * need __read_pkey_reg() version so we do not do shadow_pkey_reg
389 	 * checking
390 	 */
391 	dprintf1("signal pkey_reg from  pkey_reg: %016llx\n",
392 			__read_pkey_reg());
393 	dprintf1("pkey from siginfo: %016llx\n", siginfo_pkey);
394 #if defined(__i386__) || defined(__x86_64__) /* arch */
395 	dprintf1("signal pkey_reg from xsave: %08x\n", *pkey_reg_ptr);
396 	*(u64 *)pkey_reg_ptr = 0x00000000;
397 	dprintf1("WARNING: set PKEY_REG=0 to allow faulting instruction to continue\n");
398 #elif defined(__powerpc64__) /* arch */
399 	/* restore access and let the faulting instruction continue */
400 	pkey_access_allow(siginfo_pkey);
401 #endif /* arch */
402 	pkey_faults++;
403 	dprintf1("<<<<==================================================\n");
404 	dprint_in_signal = 0;
405 }
406 
407 int wait_all_children(void)
408 {
409 	int status;
410 	return waitpid(-1, &status, 0);
411 }
412 
413 void sig_chld(int x)
414 {
415 	dprint_in_signal = 1;
416 	dprintf2("[%d] SIGCHLD: %d\n", getpid(), x);
417 	dprint_in_signal = 0;
418 }
419 
420 void setup_sigsegv_handler(void)
421 {
422 	int r, rs;
423 	struct sigaction newact;
424 	struct sigaction oldact;
425 
426 	/* #PF is mapped to sigsegv */
427 	int signum  = SIGSEGV;
428 
429 	newact.sa_handler = 0;
430 	newact.sa_sigaction = signal_handler;
431 
432 	/*sigset_t - signals to block while in the handler */
433 	/* get the old signal mask. */
434 	rs = sigprocmask(SIG_SETMASK, 0, &newact.sa_mask);
435 	pkey_assert(rs == 0);
436 
437 	/* call sa_sigaction, not sa_handler*/
438 	newact.sa_flags = SA_SIGINFO;
439 
440 	newact.sa_restorer = 0;  /* void(*)(), obsolete */
441 	r = sigaction(signum, &newact, &oldact);
442 	r = sigaction(SIGALRM, &newact, &oldact);
443 	pkey_assert(r == 0);
444 }
445 
446 void setup_handlers(void)
447 {
448 	signal(SIGCHLD, &sig_chld);
449 	setup_sigsegv_handler();
450 }
451 
452 pid_t fork_lazy_child(void)
453 {
454 	pid_t forkret;
455 
456 	forkret = fork();
457 	pkey_assert(forkret >= 0);
458 	dprintf3("[%d] fork() ret: %d\n", getpid(), forkret);
459 
460 	if (!forkret) {
461 		/* in the child */
462 		while (1) {
463 			dprintf1("child sleeping...\n");
464 			sleep(30);
465 		}
466 	}
467 	return forkret;
468 }
469 
470 int sys_mprotect_pkey(void *ptr, size_t size, unsigned long orig_prot,
471 		unsigned long pkey)
472 {
473 	int sret;
474 
475 	dprintf2("%s(0x%p, %zx, prot=%lx, pkey=%lx)\n", __func__,
476 			ptr, size, orig_prot, pkey);
477 
478 	errno = 0;
479 	sret = syscall(SYS_mprotect_key, ptr, size, orig_prot, pkey);
480 	if (errno) {
481 		dprintf2("SYS_mprotect_key sret: %d\n", sret);
482 		dprintf2("SYS_mprotect_key prot: 0x%lx\n", orig_prot);
483 		dprintf2("SYS_mprotect_key failed, errno: %d\n", errno);
484 		if (DEBUG_LEVEL >= 2)
485 			perror("SYS_mprotect_pkey");
486 	}
487 	return sret;
488 }
489 
490 int sys_pkey_alloc(unsigned long flags, unsigned long init_val)
491 {
492 	int ret = syscall(SYS_pkey_alloc, flags, init_val);
493 	dprintf1("%s(flags=%lx, init_val=%lx) syscall ret: %d errno: %d\n",
494 			__func__, flags, init_val, ret, errno);
495 	return ret;
496 }
497 
498 int alloc_pkey(void)
499 {
500 	int ret;
501 	unsigned long init_val = 0x0;
502 
503 	dprintf1("%s()::%d, pkey_reg: 0x%016llx shadow: %016llx\n",
504 			__func__, __LINE__, __read_pkey_reg(), shadow_pkey_reg);
505 	ret = sys_pkey_alloc(0, init_val);
506 	/*
507 	 * pkey_alloc() sets PKEY register, so we need to reflect it in
508 	 * shadow_pkey_reg:
509 	 */
510 	dprintf4("%s()::%d, ret: %d pkey_reg: 0x%016llx"
511 			" shadow: 0x%016llx\n",
512 			__func__, __LINE__, ret, __read_pkey_reg(),
513 			shadow_pkey_reg);
514 	if (ret > 0) {
515 		/* clear both the bits: */
516 		shadow_pkey_reg = set_pkey_bits(shadow_pkey_reg, ret,
517 						~PKEY_MASK);
518 		dprintf4("%s()::%d, ret: %d pkey_reg: 0x%016llx"
519 				" shadow: 0x%016llx\n",
520 				__func__,
521 				__LINE__, ret, __read_pkey_reg(),
522 				shadow_pkey_reg);
523 		/*
524 		 * move the new state in from init_val
525 		 * (remember, we cheated and init_val == pkey_reg format)
526 		 */
527 		shadow_pkey_reg = set_pkey_bits(shadow_pkey_reg, ret,
528 						init_val);
529 	}
530 	dprintf4("%s()::%d, ret: %d pkey_reg: 0x%016llx"
531 			" shadow: 0x%016llx\n",
532 			__func__, __LINE__, ret, __read_pkey_reg(),
533 			shadow_pkey_reg);
534 	dprintf1("%s()::%d errno: %d\n", __func__, __LINE__, errno);
535 	/* for shadow checking: */
536 	read_pkey_reg();
537 	dprintf4("%s()::%d, ret: %d pkey_reg: 0x%016llx"
538 		 " shadow: 0x%016llx\n",
539 		__func__, __LINE__, ret, __read_pkey_reg(),
540 		shadow_pkey_reg);
541 	return ret;
542 }
543 
544 int sys_pkey_free(unsigned long pkey)
545 {
546 	int ret = syscall(SYS_pkey_free, pkey);
547 	dprintf1("%s(pkey=%ld) syscall ret: %d\n", __func__, pkey, ret);
548 	return ret;
549 }
550 
551 /*
552  * I had a bug where pkey bits could be set by mprotect() but
553  * not cleared.  This ensures we get lots of random bit sets
554  * and clears on the vma and pte pkey bits.
555  */
556 int alloc_random_pkey(void)
557 {
558 	int max_nr_pkey_allocs;
559 	int ret;
560 	int i;
561 	int alloced_pkeys[NR_PKEYS];
562 	int nr_alloced = 0;
563 	int random_index;
564 	memset(alloced_pkeys, 0, sizeof(alloced_pkeys));
565 
566 	/* allocate every possible key and make a note of which ones we got */
567 	max_nr_pkey_allocs = NR_PKEYS;
568 	for (i = 0; i < max_nr_pkey_allocs; i++) {
569 		int new_pkey = alloc_pkey();
570 		if (new_pkey < 0)
571 			break;
572 		alloced_pkeys[nr_alloced++] = new_pkey;
573 	}
574 
575 	pkey_assert(nr_alloced > 0);
576 	/* select a random one out of the allocated ones */
577 	random_index = rand() % nr_alloced;
578 	ret = alloced_pkeys[random_index];
579 	/* now zero it out so we don't free it next */
580 	alloced_pkeys[random_index] = 0;
581 
582 	/* go through the allocated ones that we did not want and free them */
583 	for (i = 0; i < nr_alloced; i++) {
584 		int free_ret;
585 		if (!alloced_pkeys[i])
586 			continue;
587 		free_ret = sys_pkey_free(alloced_pkeys[i]);
588 		pkey_assert(!free_ret);
589 	}
590 	dprintf1("%s()::%d, ret: %d pkey_reg: 0x%016llx"
591 			 " shadow: 0x%016llx\n", __func__,
592 			__LINE__, ret, __read_pkey_reg(), shadow_pkey_reg);
593 	return ret;
594 }
595 
596 int mprotect_pkey(void *ptr, size_t size, unsigned long orig_prot,
597 		unsigned long pkey)
598 {
599 	int nr_iterations = random() % 100;
600 	int ret;
601 
602 	while (0) {
603 		int rpkey = alloc_random_pkey();
604 		ret = sys_mprotect_pkey(ptr, size, orig_prot, pkey);
605 		dprintf1("sys_mprotect_pkey(%p, %zx, prot=0x%lx, pkey=%ld) ret: %d\n",
606 				ptr, size, orig_prot, pkey, ret);
607 		if (nr_iterations-- < 0)
608 			break;
609 
610 		dprintf1("%s()::%d, ret: %d pkey_reg: 0x%016llx"
611 			" shadow: 0x%016llx\n",
612 			__func__, __LINE__, ret, __read_pkey_reg(),
613 			shadow_pkey_reg);
614 		sys_pkey_free(rpkey);
615 		dprintf1("%s()::%d, ret: %d pkey_reg: 0x%016llx"
616 			" shadow: 0x%016llx\n",
617 			__func__, __LINE__, ret, __read_pkey_reg(),
618 			shadow_pkey_reg);
619 	}
620 	pkey_assert(pkey < NR_PKEYS);
621 
622 	ret = sys_mprotect_pkey(ptr, size, orig_prot, pkey);
623 	dprintf1("mprotect_pkey(%p, %zx, prot=0x%lx, pkey=%ld) ret: %d\n",
624 			ptr, size, orig_prot, pkey, ret);
625 	pkey_assert(!ret);
626 	dprintf1("%s()::%d, ret: %d pkey_reg: 0x%016llx"
627 			" shadow: 0x%016llx\n", __func__,
628 			__LINE__, ret, __read_pkey_reg(), shadow_pkey_reg);
629 	return ret;
630 }
631 
632 struct pkey_malloc_record {
633 	void *ptr;
634 	long size;
635 	int prot;
636 };
637 struct pkey_malloc_record *pkey_malloc_records;
638 struct pkey_malloc_record *pkey_last_malloc_record;
639 long nr_pkey_malloc_records;
640 void record_pkey_malloc(void *ptr, long size, int prot)
641 {
642 	long i;
643 	struct pkey_malloc_record *rec = NULL;
644 
645 	for (i = 0; i < nr_pkey_malloc_records; i++) {
646 		rec = &pkey_malloc_records[i];
647 		/* find a free record */
648 		if (rec)
649 			break;
650 	}
651 	if (!rec) {
652 		/* every record is full */
653 		size_t old_nr_records = nr_pkey_malloc_records;
654 		size_t new_nr_records = (nr_pkey_malloc_records * 2 + 1);
655 		size_t new_size = new_nr_records * sizeof(struct pkey_malloc_record);
656 		dprintf2("new_nr_records: %zd\n", new_nr_records);
657 		dprintf2("new_size: %zd\n", new_size);
658 		pkey_malloc_records = realloc(pkey_malloc_records, new_size);
659 		pkey_assert(pkey_malloc_records != NULL);
660 		rec = &pkey_malloc_records[nr_pkey_malloc_records];
661 		/*
662 		 * realloc() does not initialize memory, so zero it from
663 		 * the first new record all the way to the end.
664 		 */
665 		for (i = 0; i < new_nr_records - old_nr_records; i++)
666 			memset(rec + i, 0, sizeof(*rec));
667 	}
668 	dprintf3("filling malloc record[%d/%p]: {%p, %ld}\n",
669 		(int)(rec - pkey_malloc_records), rec, ptr, size);
670 	rec->ptr = ptr;
671 	rec->size = size;
672 	rec->prot = prot;
673 	pkey_last_malloc_record = rec;
674 	nr_pkey_malloc_records++;
675 }
676 
677 void free_pkey_malloc(void *ptr)
678 {
679 	long i;
680 	int ret;
681 	dprintf3("%s(%p)\n", __func__, ptr);
682 	for (i = 0; i < nr_pkey_malloc_records; i++) {
683 		struct pkey_malloc_record *rec = &pkey_malloc_records[i];
684 		dprintf4("looking for ptr %p at record[%ld/%p]: {%p, %ld}\n",
685 				ptr, i, rec, rec->ptr, rec->size);
686 		if ((ptr <  rec->ptr) ||
687 		    (ptr >= rec->ptr + rec->size))
688 			continue;
689 
690 		dprintf3("found ptr %p at record[%ld/%p]: {%p, %ld}\n",
691 				ptr, i, rec, rec->ptr, rec->size);
692 		nr_pkey_malloc_records--;
693 		ret = munmap(rec->ptr, rec->size);
694 		dprintf3("munmap ret: %d\n", ret);
695 		pkey_assert(!ret);
696 		dprintf3("clearing rec->ptr, rec: %p\n", rec);
697 		rec->ptr = NULL;
698 		dprintf3("done clearing rec->ptr, rec: %p\n", rec);
699 		return;
700 	}
701 	pkey_assert(false);
702 }
703 
704 
705 void *malloc_pkey_with_mprotect(long size, int prot, u16 pkey)
706 {
707 	void *ptr;
708 	int ret;
709 
710 	read_pkey_reg();
711 	dprintf1("doing %s(size=%ld, prot=0x%x, pkey=%d)\n", __func__,
712 			size, prot, pkey);
713 	pkey_assert(pkey < NR_PKEYS);
714 	ptr = mmap(NULL, size, prot, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
715 	pkey_assert(ptr != (void *)-1);
716 	ret = mprotect_pkey((void *)ptr, PAGE_SIZE, prot, pkey);
717 	pkey_assert(!ret);
718 	record_pkey_malloc(ptr, size, prot);
719 	read_pkey_reg();
720 
721 	dprintf1("%s() for pkey %d @ %p\n", __func__, pkey, ptr);
722 	return ptr;
723 }
724 
725 void *malloc_pkey_anon_huge(long size, int prot, u16 pkey)
726 {
727 	int ret;
728 	void *ptr;
729 
730 	dprintf1("doing %s(size=%ld, prot=0x%x, pkey=%d)\n", __func__,
731 			size, prot, pkey);
732 	/*
733 	 * Guarantee we can fit at least one huge page in the resulting
734 	 * allocation by allocating space for 2:
735 	 */
736 	size = ALIGN_UP(size, HPAGE_SIZE * 2);
737 	ptr = mmap(NULL, size, PROT_NONE, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
738 	pkey_assert(ptr != (void *)-1);
739 	record_pkey_malloc(ptr, size, prot);
740 	mprotect_pkey(ptr, size, prot, pkey);
741 
742 	dprintf1("unaligned ptr: %p\n", ptr);
743 	ptr = ALIGN_PTR_UP(ptr, HPAGE_SIZE);
744 	dprintf1("  aligned ptr: %p\n", ptr);
745 	ret = madvise(ptr, HPAGE_SIZE, MADV_HUGEPAGE);
746 	dprintf1("MADV_HUGEPAGE ret: %d\n", ret);
747 	ret = madvise(ptr, HPAGE_SIZE, MADV_WILLNEED);
748 	dprintf1("MADV_WILLNEED ret: %d\n", ret);
749 	memset(ptr, 0, HPAGE_SIZE);
750 
751 	dprintf1("mmap()'d thp for pkey %d @ %p\n", pkey, ptr);
752 	return ptr;
753 }
754 
755 int hugetlb_setup_ok;
756 #define SYSFS_FMT_NR_HUGE_PAGES "/sys/kernel/mm/hugepages/hugepages-%ldkB/nr_hugepages"
757 #define GET_NR_HUGE_PAGES 10
758 void setup_hugetlbfs(void)
759 {
760 	int err;
761 	int fd;
762 	char buf[256];
763 	long hpagesz_kb;
764 	long hpagesz_mb;
765 
766 	if (geteuid() != 0) {
767 		fprintf(stderr, "WARNING: not run as root, can not do hugetlb test\n");
768 		return;
769 	}
770 
771 	cat_into_file(__stringify(GET_NR_HUGE_PAGES), "/proc/sys/vm/nr_hugepages");
772 
773 	/*
774 	 * Now go make sure that we got the pages and that they
775 	 * are PMD-level pages. Someone might have made PUD-level
776 	 * pages the default.
777 	 */
778 	hpagesz_kb = HPAGE_SIZE / 1024;
779 	hpagesz_mb = hpagesz_kb / 1024;
780 	sprintf(buf, SYSFS_FMT_NR_HUGE_PAGES, hpagesz_kb);
781 	fd = open(buf, O_RDONLY);
782 	if (fd < 0) {
783 		fprintf(stderr, "opening sysfs %ldM hugetlb config: %s\n",
784 			hpagesz_mb, strerror(errno));
785 		return;
786 	}
787 
788 	/* -1 to guarantee leaving the trailing \0 */
789 	err = read(fd, buf, sizeof(buf)-1);
790 	close(fd);
791 	if (err <= 0) {
792 		fprintf(stderr, "reading sysfs %ldM hugetlb config: %s\n",
793 			hpagesz_mb, strerror(errno));
794 		return;
795 	}
796 
797 	if (atoi(buf) != GET_NR_HUGE_PAGES) {
798 		fprintf(stderr, "could not confirm %ldM pages, got: '%s' expected %d\n",
799 			hpagesz_mb, buf, GET_NR_HUGE_PAGES);
800 		return;
801 	}
802 
803 	hugetlb_setup_ok = 1;
804 }
805 
806 void *malloc_pkey_hugetlb(long size, int prot, u16 pkey)
807 {
808 	void *ptr;
809 	int flags = MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB;
810 
811 	if (!hugetlb_setup_ok)
812 		return PTR_ERR_ENOTSUP;
813 
814 	dprintf1("doing %s(%ld, %x, %x)\n", __func__, size, prot, pkey);
815 	size = ALIGN_UP(size, HPAGE_SIZE * 2);
816 	pkey_assert(pkey < NR_PKEYS);
817 	ptr = mmap(NULL, size, PROT_NONE, flags, -1, 0);
818 	pkey_assert(ptr != (void *)-1);
819 	mprotect_pkey(ptr, size, prot, pkey);
820 
821 	record_pkey_malloc(ptr, size, prot);
822 
823 	dprintf1("mmap()'d hugetlbfs for pkey %d @ %p\n", pkey, ptr);
824 	return ptr;
825 }
826 
827 void *malloc_pkey_mmap_dax(long size, int prot, u16 pkey)
828 {
829 	void *ptr;
830 	int fd;
831 
832 	dprintf1("doing %s(size=%ld, prot=0x%x, pkey=%d)\n", __func__,
833 			size, prot, pkey);
834 	pkey_assert(pkey < NR_PKEYS);
835 	fd = open("/dax/foo", O_RDWR);
836 	pkey_assert(fd >= 0);
837 
838 	ptr = mmap(0, size, prot, MAP_SHARED, fd, 0);
839 	pkey_assert(ptr != (void *)-1);
840 
841 	mprotect_pkey(ptr, size, prot, pkey);
842 
843 	record_pkey_malloc(ptr, size, prot);
844 
845 	dprintf1("mmap()'d for pkey %d @ %p\n", pkey, ptr);
846 	close(fd);
847 	return ptr;
848 }
849 
850 void *(*pkey_malloc[])(long size, int prot, u16 pkey) = {
851 
852 	malloc_pkey_with_mprotect,
853 	malloc_pkey_with_mprotect_subpage,
854 	malloc_pkey_anon_huge,
855 	malloc_pkey_hugetlb
856 /* can not do direct with the pkey_mprotect() API:
857 	malloc_pkey_mmap_direct,
858 	malloc_pkey_mmap_dax,
859 */
860 };
861 
862 void *malloc_pkey(long size, int prot, u16 pkey)
863 {
864 	void *ret;
865 	static int malloc_type;
866 	int nr_malloc_types = ARRAY_SIZE(pkey_malloc);
867 
868 	pkey_assert(pkey < NR_PKEYS);
869 
870 	while (1) {
871 		pkey_assert(malloc_type < nr_malloc_types);
872 
873 		ret = pkey_malloc[malloc_type](size, prot, pkey);
874 		pkey_assert(ret != (void *)-1);
875 
876 		malloc_type++;
877 		if (malloc_type >= nr_malloc_types)
878 			malloc_type = (random()%nr_malloc_types);
879 
880 		/* try again if the malloc_type we tried is unsupported */
881 		if (ret == PTR_ERR_ENOTSUP)
882 			continue;
883 
884 		break;
885 	}
886 
887 	dprintf3("%s(%ld, prot=%x, pkey=%x) returning: %p\n", __func__,
888 			size, prot, pkey, ret);
889 	return ret;
890 }
891 
892 int last_pkey_faults;
893 #define UNKNOWN_PKEY -2
894 void expected_pkey_fault(int pkey)
895 {
896 	dprintf2("%s(): last_pkey_faults: %d pkey_faults: %d\n",
897 			__func__, last_pkey_faults, pkey_faults);
898 	dprintf2("%s(%d): last_si_pkey: %d\n", __func__, pkey, last_si_pkey);
899 	pkey_assert(last_pkey_faults + 1 == pkey_faults);
900 
901        /*
902 	* For exec-only memory, we do not know the pkey in
903 	* advance, so skip this check.
904 	*/
905 	if (pkey != UNKNOWN_PKEY)
906 		pkey_assert(last_si_pkey == pkey);
907 
908 #if defined(__i386__) || defined(__x86_64__) /* arch */
909 	/*
910 	 * The signal handler shold have cleared out PKEY register to let the
911 	 * test program continue.  We now have to restore it.
912 	 */
913 	if (__read_pkey_reg() != 0)
914 #else /* arch */
915 	if (__read_pkey_reg() != shadow_pkey_reg)
916 #endif /* arch */
917 		pkey_assert(0);
918 
919 	__write_pkey_reg(shadow_pkey_reg);
920 	dprintf1("%s() set pkey_reg=%016llx to restore state after signal "
921 		       "nuked it\n", __func__, shadow_pkey_reg);
922 	last_pkey_faults = pkey_faults;
923 	last_si_pkey = -1;
924 }
925 
926 #define do_not_expect_pkey_fault(msg)	do {			\
927 	if (last_pkey_faults != pkey_faults)			\
928 		dprintf0("unexpected PKey fault: %s\n", msg);	\
929 	pkey_assert(last_pkey_faults == pkey_faults);		\
930 } while (0)
931 
932 int test_fds[10] = { -1 };
933 int nr_test_fds;
934 void __save_test_fd(int fd)
935 {
936 	pkey_assert(fd >= 0);
937 	pkey_assert(nr_test_fds < ARRAY_SIZE(test_fds));
938 	test_fds[nr_test_fds] = fd;
939 	nr_test_fds++;
940 }
941 
942 int get_test_read_fd(void)
943 {
944 	int test_fd = open("/etc/passwd", O_RDONLY);
945 	__save_test_fd(test_fd);
946 	return test_fd;
947 }
948 
949 void close_test_fds(void)
950 {
951 	int i;
952 
953 	for (i = 0; i < nr_test_fds; i++) {
954 		if (test_fds[i] < 0)
955 			continue;
956 		close(test_fds[i]);
957 		test_fds[i] = -1;
958 	}
959 	nr_test_fds = 0;
960 }
961 
962 #define barrier() __asm__ __volatile__("": : :"memory")
963 __attribute__((noinline)) int read_ptr(int *ptr)
964 {
965 	/*
966 	 * Keep GCC from optimizing this away somehow
967 	 */
968 	barrier();
969 	return *ptr;
970 }
971 
972 void test_pkey_alloc_free_attach_pkey0(int *ptr, u16 pkey)
973 {
974 	int i, err;
975 	int max_nr_pkey_allocs;
976 	int alloced_pkeys[NR_PKEYS];
977 	int nr_alloced = 0;
978 	long size;
979 
980 	pkey_assert(pkey_last_malloc_record);
981 	size = pkey_last_malloc_record->size;
982 	/*
983 	 * This is a bit of a hack.  But mprotect() requires
984 	 * huge-page-aligned sizes when operating on hugetlbfs.
985 	 * So, make sure that we use something that's a multiple
986 	 * of a huge page when we can.
987 	 */
988 	if (size >= HPAGE_SIZE)
989 		size = HPAGE_SIZE;
990 
991 	/* allocate every possible key and make sure key-0 never got allocated */
992 	max_nr_pkey_allocs = NR_PKEYS;
993 	for (i = 0; i < max_nr_pkey_allocs; i++) {
994 		int new_pkey = alloc_pkey();
995 		pkey_assert(new_pkey != 0);
996 
997 		if (new_pkey < 0)
998 			break;
999 		alloced_pkeys[nr_alloced++] = new_pkey;
1000 	}
1001 	/* free all the allocated keys */
1002 	for (i = 0; i < nr_alloced; i++) {
1003 		int free_ret;
1004 
1005 		if (!alloced_pkeys[i])
1006 			continue;
1007 		free_ret = sys_pkey_free(alloced_pkeys[i]);
1008 		pkey_assert(!free_ret);
1009 	}
1010 
1011 	/* attach key-0 in various modes */
1012 	err = sys_mprotect_pkey(ptr, size, PROT_READ, 0);
1013 	pkey_assert(!err);
1014 	err = sys_mprotect_pkey(ptr, size, PROT_WRITE, 0);
1015 	pkey_assert(!err);
1016 	err = sys_mprotect_pkey(ptr, size, PROT_EXEC, 0);
1017 	pkey_assert(!err);
1018 	err = sys_mprotect_pkey(ptr, size, PROT_READ|PROT_WRITE, 0);
1019 	pkey_assert(!err);
1020 	err = sys_mprotect_pkey(ptr, size, PROT_READ|PROT_WRITE|PROT_EXEC, 0);
1021 	pkey_assert(!err);
1022 }
1023 
1024 void test_read_of_write_disabled_region(int *ptr, u16 pkey)
1025 {
1026 	int ptr_contents;
1027 
1028 	dprintf1("disabling write access to PKEY[1], doing read\n");
1029 	pkey_write_deny(pkey);
1030 	ptr_contents = read_ptr(ptr);
1031 	dprintf1("*ptr: %d\n", ptr_contents);
1032 	dprintf1("\n");
1033 }
1034 void test_read_of_access_disabled_region(int *ptr, u16 pkey)
1035 {
1036 	int ptr_contents;
1037 
1038 	dprintf1("disabling access to PKEY[%02d], doing read @ %p\n", pkey, ptr);
1039 	read_pkey_reg();
1040 	pkey_access_deny(pkey);
1041 	ptr_contents = read_ptr(ptr);
1042 	dprintf1("*ptr: %d\n", ptr_contents);
1043 	expected_pkey_fault(pkey);
1044 }
1045 
1046 void test_read_of_access_disabled_region_with_page_already_mapped(int *ptr,
1047 		u16 pkey)
1048 {
1049 	int ptr_contents;
1050 
1051 	dprintf1("disabling access to PKEY[%02d], doing read @ %p\n",
1052 				pkey, ptr);
1053 	ptr_contents = read_ptr(ptr);
1054 	dprintf1("reading ptr before disabling the read : %d\n",
1055 			ptr_contents);
1056 	read_pkey_reg();
1057 	pkey_access_deny(pkey);
1058 	ptr_contents = read_ptr(ptr);
1059 	dprintf1("*ptr: %d\n", ptr_contents);
1060 	expected_pkey_fault(pkey);
1061 }
1062 
1063 void test_write_of_write_disabled_region_with_page_already_mapped(int *ptr,
1064 		u16 pkey)
1065 {
1066 	*ptr = __LINE__;
1067 	dprintf1("disabling write access; after accessing the page, "
1068 		"to PKEY[%02d], doing write\n", pkey);
1069 	pkey_write_deny(pkey);
1070 	*ptr = __LINE__;
1071 	expected_pkey_fault(pkey);
1072 }
1073 
1074 void test_write_of_write_disabled_region(int *ptr, u16 pkey)
1075 {
1076 	dprintf1("disabling write access to PKEY[%02d], doing write\n", pkey);
1077 	pkey_write_deny(pkey);
1078 	*ptr = __LINE__;
1079 	expected_pkey_fault(pkey);
1080 }
1081 void test_write_of_access_disabled_region(int *ptr, u16 pkey)
1082 {
1083 	dprintf1("disabling access to PKEY[%02d], doing write\n", pkey);
1084 	pkey_access_deny(pkey);
1085 	*ptr = __LINE__;
1086 	expected_pkey_fault(pkey);
1087 }
1088 
1089 void test_write_of_access_disabled_region_with_page_already_mapped(int *ptr,
1090 			u16 pkey)
1091 {
1092 	*ptr = __LINE__;
1093 	dprintf1("disabling access; after accessing the page, "
1094 		" to PKEY[%02d], doing write\n", pkey);
1095 	pkey_access_deny(pkey);
1096 	*ptr = __LINE__;
1097 	expected_pkey_fault(pkey);
1098 }
1099 
1100 void test_kernel_write_of_access_disabled_region(int *ptr, u16 pkey)
1101 {
1102 	int ret;
1103 	int test_fd = get_test_read_fd();
1104 
1105 	dprintf1("disabling access to PKEY[%02d], "
1106 		 "having kernel read() to buffer\n", pkey);
1107 	pkey_access_deny(pkey);
1108 	ret = read(test_fd, ptr, 1);
1109 	dprintf1("read ret: %d\n", ret);
1110 	pkey_assert(ret);
1111 }
1112 void test_kernel_write_of_write_disabled_region(int *ptr, u16 pkey)
1113 {
1114 	int ret;
1115 	int test_fd = get_test_read_fd();
1116 
1117 	pkey_write_deny(pkey);
1118 	ret = read(test_fd, ptr, 100);
1119 	dprintf1("read ret: %d\n", ret);
1120 	if (ret < 0 && (DEBUG_LEVEL > 0))
1121 		perror("verbose read result (OK for this to be bad)");
1122 	pkey_assert(ret);
1123 }
1124 
1125 void test_kernel_gup_of_access_disabled_region(int *ptr, u16 pkey)
1126 {
1127 	int pipe_ret, vmsplice_ret;
1128 	struct iovec iov;
1129 	int pipe_fds[2];
1130 
1131 	pipe_ret = pipe(pipe_fds);
1132 
1133 	pkey_assert(pipe_ret == 0);
1134 	dprintf1("disabling access to PKEY[%02d], "
1135 		 "having kernel vmsplice from buffer\n", pkey);
1136 	pkey_access_deny(pkey);
1137 	iov.iov_base = ptr;
1138 	iov.iov_len = PAGE_SIZE;
1139 	vmsplice_ret = vmsplice(pipe_fds[1], &iov, 1, SPLICE_F_GIFT);
1140 	dprintf1("vmsplice() ret: %d\n", vmsplice_ret);
1141 	pkey_assert(vmsplice_ret == -1);
1142 
1143 	close(pipe_fds[0]);
1144 	close(pipe_fds[1]);
1145 }
1146 
1147 void test_kernel_gup_write_to_write_disabled_region(int *ptr, u16 pkey)
1148 {
1149 	int ignored = 0xdada;
1150 	int futex_ret;
1151 	int some_int = __LINE__;
1152 
1153 	dprintf1("disabling write to PKEY[%02d], "
1154 		 "doing futex gunk in buffer\n", pkey);
1155 	*ptr = some_int;
1156 	pkey_write_deny(pkey);
1157 	futex_ret = syscall(SYS_futex, ptr, FUTEX_WAIT, some_int-1, NULL,
1158 			&ignored, ignored);
1159 	if (DEBUG_LEVEL > 0)
1160 		perror("futex");
1161 	dprintf1("futex() ret: %d\n", futex_ret);
1162 }
1163 
1164 /* Assumes that all pkeys other than 'pkey' are unallocated */
1165 void test_pkey_syscalls_on_non_allocated_pkey(int *ptr, u16 pkey)
1166 {
1167 	int err;
1168 	int i;
1169 
1170 	/* Note: 0 is the default pkey, so don't mess with it */
1171 	for (i = 1; i < NR_PKEYS; i++) {
1172 		if (pkey == i)
1173 			continue;
1174 
1175 		dprintf1("trying get/set/free to non-allocated pkey: %2d\n", i);
1176 		err = sys_pkey_free(i);
1177 		pkey_assert(err);
1178 
1179 		err = sys_pkey_free(i);
1180 		pkey_assert(err);
1181 
1182 		err = sys_mprotect_pkey(ptr, PAGE_SIZE, PROT_READ, i);
1183 		pkey_assert(err);
1184 	}
1185 }
1186 
1187 /* Assumes that all pkeys other than 'pkey' are unallocated */
1188 void test_pkey_syscalls_bad_args(int *ptr, u16 pkey)
1189 {
1190 	int err;
1191 	int bad_pkey = NR_PKEYS+99;
1192 
1193 	/* pass a known-invalid pkey in: */
1194 	err = sys_mprotect_pkey(ptr, PAGE_SIZE, PROT_READ, bad_pkey);
1195 	pkey_assert(err);
1196 }
1197 
1198 void become_child(void)
1199 {
1200 	pid_t forkret;
1201 
1202 	forkret = fork();
1203 	pkey_assert(forkret >= 0);
1204 	dprintf3("[%d] fork() ret: %d\n", getpid(), forkret);
1205 
1206 	if (!forkret) {
1207 		/* in the child */
1208 		return;
1209 	}
1210 	exit(0);
1211 }
1212 
1213 /* Assumes that all pkeys other than 'pkey' are unallocated */
1214 void test_pkey_alloc_exhaust(int *ptr, u16 pkey)
1215 {
1216 	int err;
1217 	int allocated_pkeys[NR_PKEYS] = {0};
1218 	int nr_allocated_pkeys = 0;
1219 	int i;
1220 
1221 	for (i = 0; i < NR_PKEYS*3; i++) {
1222 		int new_pkey;
1223 		dprintf1("%s() alloc loop: %d\n", __func__, i);
1224 		new_pkey = alloc_pkey();
1225 		dprintf4("%s()::%d, err: %d pkey_reg: 0x%016llx"
1226 				" shadow: 0x%016llx\n",
1227 				__func__, __LINE__, err, __read_pkey_reg(),
1228 				shadow_pkey_reg);
1229 		read_pkey_reg(); /* for shadow checking */
1230 		dprintf2("%s() errno: %d ENOSPC: %d\n", __func__, errno, ENOSPC);
1231 		if ((new_pkey == -1) && (errno == ENOSPC)) {
1232 			dprintf2("%s() failed to allocate pkey after %d tries\n",
1233 				__func__, nr_allocated_pkeys);
1234 		} else {
1235 			/*
1236 			 * Ensure the number of successes never
1237 			 * exceeds the number of keys supported
1238 			 * in the hardware.
1239 			 */
1240 			pkey_assert(nr_allocated_pkeys < NR_PKEYS);
1241 			allocated_pkeys[nr_allocated_pkeys++] = new_pkey;
1242 		}
1243 
1244 		/*
1245 		 * Make sure that allocation state is properly
1246 		 * preserved across fork().
1247 		 */
1248 		if (i == NR_PKEYS*2)
1249 			become_child();
1250 	}
1251 
1252 	dprintf3("%s()::%d\n", __func__, __LINE__);
1253 
1254 	/*
1255 	 * On x86:
1256 	 * There are 16 pkeys supported in hardware.  Three are
1257 	 * allocated by the time we get here:
1258 	 *   1. The default key (0)
1259 	 *   2. One possibly consumed by an execute-only mapping.
1260 	 *   3. One allocated by the test code and passed in via
1261 	 *      'pkey' to this function.
1262 	 * Ensure that we can allocate at least another 13 (16-3).
1263 	 *
1264 	 * On powerpc:
1265 	 * There are either 5, 28, 29 or 32 pkeys supported in
1266 	 * hardware depending on the page size (4K or 64K) and
1267 	 * platform (powernv or powervm). Four are allocated by
1268 	 * the time we get here. These include pkey-0, pkey-1,
1269 	 * exec-only pkey and the one allocated by the test code.
1270 	 * Ensure that we can allocate the remaining.
1271 	 */
1272 	pkey_assert(i >= (NR_PKEYS - get_arch_reserved_keys() - 1));
1273 
1274 	for (i = 0; i < nr_allocated_pkeys; i++) {
1275 		err = sys_pkey_free(allocated_pkeys[i]);
1276 		pkey_assert(!err);
1277 		read_pkey_reg(); /* for shadow checking */
1278 	}
1279 }
1280 
1281 void arch_force_pkey_reg_init(void)
1282 {
1283 #if defined(__i386__) || defined(__x86_64__) /* arch */
1284 	u64 *buf;
1285 
1286 	/*
1287 	 * All keys should be allocated and set to allow reads and
1288 	 * writes, so the register should be all 0.  If not, just
1289 	 * skip the test.
1290 	 */
1291 	if (read_pkey_reg())
1292 		return;
1293 
1294 	/*
1295 	 * Just allocate an absurd about of memory rather than
1296 	 * doing the XSAVE size enumeration dance.
1297 	 */
1298 	buf = mmap(NULL, 1*MB, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
1299 
1300 	/* These __builtins require compiling with -mxsave */
1301 
1302 	/* XSAVE to build a valid buffer: */
1303 	__builtin_ia32_xsave(buf, XSTATE_PKEY);
1304 	/* Clear XSTATE_BV[PKRU]: */
1305 	buf[XSTATE_BV_OFFSET/sizeof(u64)] &= ~XSTATE_PKEY;
1306 	/* XRSTOR will likely get PKRU back to the init state: */
1307 	__builtin_ia32_xrstor(buf, XSTATE_PKEY);
1308 
1309 	munmap(buf, 1*MB);
1310 #endif
1311 }
1312 
1313 
1314 /*
1315  * This is mostly useless on ppc for now.  But it will not
1316  * hurt anything and should give some better coverage as
1317  * a long-running test that continually checks the pkey
1318  * register.
1319  */
1320 void test_pkey_init_state(int *ptr, u16 pkey)
1321 {
1322 	int err;
1323 	int allocated_pkeys[NR_PKEYS] = {0};
1324 	int nr_allocated_pkeys = 0;
1325 	int i;
1326 
1327 	for (i = 0; i < NR_PKEYS; i++) {
1328 		int new_pkey = alloc_pkey();
1329 
1330 		if (new_pkey < 0)
1331 			continue;
1332 		allocated_pkeys[nr_allocated_pkeys++] = new_pkey;
1333 	}
1334 
1335 	dprintf3("%s()::%d\n", __func__, __LINE__);
1336 
1337 	arch_force_pkey_reg_init();
1338 
1339 	/*
1340 	 * Loop for a bit, hoping to get exercise the kernel
1341 	 * context switch code.
1342 	 */
1343 	for (i = 0; i < 1000000; i++)
1344 		read_pkey_reg();
1345 
1346 	for (i = 0; i < nr_allocated_pkeys; i++) {
1347 		err = sys_pkey_free(allocated_pkeys[i]);
1348 		pkey_assert(!err);
1349 		read_pkey_reg(); /* for shadow checking */
1350 	}
1351 }
1352 
1353 /*
1354  * pkey 0 is special.  It is allocated by default, so you do not
1355  * have to call pkey_alloc() to use it first.  Make sure that it
1356  * is usable.
1357  */
1358 void test_mprotect_with_pkey_0(int *ptr, u16 pkey)
1359 {
1360 	long size;
1361 	int prot;
1362 
1363 	assert(pkey_last_malloc_record);
1364 	size = pkey_last_malloc_record->size;
1365 	/*
1366 	 * This is a bit of a hack.  But mprotect() requires
1367 	 * huge-page-aligned sizes when operating on hugetlbfs.
1368 	 * So, make sure that we use something that's a multiple
1369 	 * of a huge page when we can.
1370 	 */
1371 	if (size >= HPAGE_SIZE)
1372 		size = HPAGE_SIZE;
1373 	prot = pkey_last_malloc_record->prot;
1374 
1375 	/* Use pkey 0 */
1376 	mprotect_pkey(ptr, size, prot, 0);
1377 
1378 	/* Make sure that we can set it back to the original pkey. */
1379 	mprotect_pkey(ptr, size, prot, pkey);
1380 }
1381 
1382 void test_ptrace_of_child(int *ptr, u16 pkey)
1383 {
1384 	__attribute__((__unused__)) int peek_result;
1385 	pid_t child_pid;
1386 	void *ignored = 0;
1387 	long ret;
1388 	int status;
1389 	/*
1390 	 * This is the "control" for our little expermient.  Make sure
1391 	 * we can always access it when ptracing.
1392 	 */
1393 	int *plain_ptr_unaligned = malloc(HPAGE_SIZE);
1394 	int *plain_ptr = ALIGN_PTR_UP(plain_ptr_unaligned, PAGE_SIZE);
1395 
1396 	/*
1397 	 * Fork a child which is an exact copy of this process, of course.
1398 	 * That means we can do all of our tests via ptrace() and then plain
1399 	 * memory access and ensure they work differently.
1400 	 */
1401 	child_pid = fork_lazy_child();
1402 	dprintf1("[%d] child pid: %d\n", getpid(), child_pid);
1403 
1404 	ret = ptrace(PTRACE_ATTACH, child_pid, ignored, ignored);
1405 	if (ret)
1406 		perror("attach");
1407 	dprintf1("[%d] attach ret: %ld %d\n", getpid(), ret, __LINE__);
1408 	pkey_assert(ret != -1);
1409 	ret = waitpid(child_pid, &status, WUNTRACED);
1410 	if ((ret != child_pid) || !(WIFSTOPPED(status))) {
1411 		fprintf(stderr, "weird waitpid result %ld stat %x\n",
1412 				ret, status);
1413 		pkey_assert(0);
1414 	}
1415 	dprintf2("waitpid ret: %ld\n", ret);
1416 	dprintf2("waitpid status: %d\n", status);
1417 
1418 	pkey_access_deny(pkey);
1419 	pkey_write_deny(pkey);
1420 
1421 	/* Write access, untested for now:
1422 	ret = ptrace(PTRACE_POKEDATA, child_pid, peek_at, data);
1423 	pkey_assert(ret != -1);
1424 	dprintf1("poke at %p: %ld\n", peek_at, ret);
1425 	*/
1426 
1427 	/*
1428 	 * Try to access the pkey-protected "ptr" via ptrace:
1429 	 */
1430 	ret = ptrace(PTRACE_PEEKDATA, child_pid, ptr, ignored);
1431 	/* expect it to work, without an error: */
1432 	pkey_assert(ret != -1);
1433 	/* Now access from the current task, and expect an exception: */
1434 	peek_result = read_ptr(ptr);
1435 	expected_pkey_fault(pkey);
1436 
1437 	/*
1438 	 * Try to access the NON-pkey-protected "plain_ptr" via ptrace:
1439 	 */
1440 	ret = ptrace(PTRACE_PEEKDATA, child_pid, plain_ptr, ignored);
1441 	/* expect it to work, without an error: */
1442 	pkey_assert(ret != -1);
1443 	/* Now access from the current task, and expect NO exception: */
1444 	peek_result = read_ptr(plain_ptr);
1445 	do_not_expect_pkey_fault("read plain pointer after ptrace");
1446 
1447 	ret = ptrace(PTRACE_DETACH, child_pid, ignored, 0);
1448 	pkey_assert(ret != -1);
1449 
1450 	ret = kill(child_pid, SIGKILL);
1451 	pkey_assert(ret != -1);
1452 
1453 	wait(&status);
1454 
1455 	free(plain_ptr_unaligned);
1456 }
1457 
1458 void *get_pointer_to_instructions(void)
1459 {
1460 	void *p1;
1461 
1462 	p1 = ALIGN_PTR_UP(&lots_o_noops_around_write, PAGE_SIZE);
1463 	dprintf3("&lots_o_noops: %p\n", &lots_o_noops_around_write);
1464 	/* lots_o_noops_around_write should be page-aligned already */
1465 	assert(p1 == &lots_o_noops_around_write);
1466 
1467 	/* Point 'p1' at the *second* page of the function: */
1468 	p1 += PAGE_SIZE;
1469 
1470 	/*
1471 	 * Try to ensure we fault this in on next touch to ensure
1472 	 * we get an instruction fault as opposed to a data one
1473 	 */
1474 	madvise(p1, PAGE_SIZE, MADV_DONTNEED);
1475 
1476 	return p1;
1477 }
1478 
1479 void test_executing_on_unreadable_memory(int *ptr, u16 pkey)
1480 {
1481 	void *p1;
1482 	int scratch;
1483 	int ptr_contents;
1484 	int ret;
1485 
1486 	p1 = get_pointer_to_instructions();
1487 	lots_o_noops_around_write(&scratch);
1488 	ptr_contents = read_ptr(p1);
1489 	dprintf2("ptr (%p) contents@%d: %x\n", p1, __LINE__, ptr_contents);
1490 
1491 	ret = mprotect_pkey(p1, PAGE_SIZE, PROT_EXEC, (u64)pkey);
1492 	pkey_assert(!ret);
1493 	pkey_access_deny(pkey);
1494 
1495 	dprintf2("pkey_reg: %016llx\n", read_pkey_reg());
1496 
1497 	/*
1498 	 * Make sure this is an *instruction* fault
1499 	 */
1500 	madvise(p1, PAGE_SIZE, MADV_DONTNEED);
1501 	lots_o_noops_around_write(&scratch);
1502 	do_not_expect_pkey_fault("executing on PROT_EXEC memory");
1503 	expect_fault_on_read_execonly_key(p1, pkey);
1504 }
1505 
1506 void test_implicit_mprotect_exec_only_memory(int *ptr, u16 pkey)
1507 {
1508 	void *p1;
1509 	int scratch;
1510 	int ptr_contents;
1511 	int ret;
1512 
1513 	dprintf1("%s() start\n", __func__);
1514 
1515 	p1 = get_pointer_to_instructions();
1516 	lots_o_noops_around_write(&scratch);
1517 	ptr_contents = read_ptr(p1);
1518 	dprintf2("ptr (%p) contents@%d: %x\n", p1, __LINE__, ptr_contents);
1519 
1520 	/* Use a *normal* mprotect(), not mprotect_pkey(): */
1521 	ret = mprotect(p1, PAGE_SIZE, PROT_EXEC);
1522 	pkey_assert(!ret);
1523 
1524 	/*
1525 	 * Reset the shadow, assuming that the above mprotect()
1526 	 * correctly changed PKRU, but to an unknown value since
1527 	 * the actual allocated pkey is unknown.
1528 	 */
1529 	shadow_pkey_reg = __read_pkey_reg();
1530 
1531 	dprintf2("pkey_reg: %016llx\n", read_pkey_reg());
1532 
1533 	/* Make sure this is an *instruction* fault */
1534 	madvise(p1, PAGE_SIZE, MADV_DONTNEED);
1535 	lots_o_noops_around_write(&scratch);
1536 	do_not_expect_pkey_fault("executing on PROT_EXEC memory");
1537 	expect_fault_on_read_execonly_key(p1, UNKNOWN_PKEY);
1538 
1539 	/*
1540 	 * Put the memory back to non-PROT_EXEC.  Should clear the
1541 	 * exec-only pkey off the VMA and allow it to be readable
1542 	 * again.  Go to PROT_NONE first to check for a kernel bug
1543 	 * that did not clear the pkey when doing PROT_NONE.
1544 	 */
1545 	ret = mprotect(p1, PAGE_SIZE, PROT_NONE);
1546 	pkey_assert(!ret);
1547 
1548 	ret = mprotect(p1, PAGE_SIZE, PROT_READ|PROT_EXEC);
1549 	pkey_assert(!ret);
1550 	ptr_contents = read_ptr(p1);
1551 	do_not_expect_pkey_fault("plain read on recently PROT_EXEC area");
1552 }
1553 
1554 #if defined(__i386__) || defined(__x86_64__)
1555 void test_ptrace_modifies_pkru(int *ptr, u16 pkey)
1556 {
1557 	u32 new_pkru;
1558 	pid_t child;
1559 	int status, ret;
1560 	int pkey_offset = pkey_reg_xstate_offset();
1561 	size_t xsave_size = cpu_max_xsave_size();
1562 	void *xsave;
1563 	u32 *pkey_register;
1564 	u64 *xstate_bv;
1565 	struct iovec iov;
1566 
1567 	new_pkru = ~read_pkey_reg();
1568 	/* Don't make PROT_EXEC mappings inaccessible */
1569 	new_pkru &= ~3;
1570 
1571 	child = fork();
1572 	pkey_assert(child >= 0);
1573 	dprintf3("[%d] fork() ret: %d\n", getpid(), child);
1574 	if (!child) {
1575 		ptrace(PTRACE_TRACEME, 0, 0, 0);
1576 		/* Stop and allow the tracer to modify PKRU directly */
1577 		raise(SIGSTOP);
1578 
1579 		/*
1580 		 * need __read_pkey_reg() version so we do not do shadow_pkey_reg
1581 		 * checking
1582 		 */
1583 		if (__read_pkey_reg() != new_pkru)
1584 			exit(1);
1585 
1586 		/* Stop and allow the tracer to clear XSTATE_BV for PKRU */
1587 		raise(SIGSTOP);
1588 
1589 		if (__read_pkey_reg() != 0)
1590 			exit(1);
1591 
1592 		/* Stop and allow the tracer to examine PKRU */
1593 		raise(SIGSTOP);
1594 
1595 		exit(0);
1596 	}
1597 
1598 	pkey_assert(child == waitpid(child, &status, 0));
1599 	dprintf3("[%d] waitpid(%d) status: %x\n", getpid(), child, status);
1600 	pkey_assert(WIFSTOPPED(status) && WSTOPSIG(status) == SIGSTOP);
1601 
1602 	xsave = (void *)malloc(xsave_size);
1603 	pkey_assert(xsave > 0);
1604 
1605 	/* Modify the PKRU register directly */
1606 	iov.iov_base = xsave;
1607 	iov.iov_len = xsave_size;
1608 	ret = ptrace(PTRACE_GETREGSET, child, (void *)NT_X86_XSTATE, &iov);
1609 	pkey_assert(ret == 0);
1610 
1611 	pkey_register = (u32 *)(xsave + pkey_offset);
1612 	pkey_assert(*pkey_register == read_pkey_reg());
1613 
1614 	*pkey_register = new_pkru;
1615 
1616 	ret = ptrace(PTRACE_SETREGSET, child, (void *)NT_X86_XSTATE, &iov);
1617 	pkey_assert(ret == 0);
1618 
1619 	/* Test that the modification is visible in ptrace before any execution */
1620 	memset(xsave, 0xCC, xsave_size);
1621 	ret = ptrace(PTRACE_GETREGSET, child, (void *)NT_X86_XSTATE, &iov);
1622 	pkey_assert(ret == 0);
1623 	pkey_assert(*pkey_register == new_pkru);
1624 
1625 	/* Execute the tracee */
1626 	ret = ptrace(PTRACE_CONT, child, 0, 0);
1627 	pkey_assert(ret == 0);
1628 
1629 	/* Test that the tracee saw the PKRU value change */
1630 	pkey_assert(child == waitpid(child, &status, 0));
1631 	dprintf3("[%d] waitpid(%d) status: %x\n", getpid(), child, status);
1632 	pkey_assert(WIFSTOPPED(status) && WSTOPSIG(status) == SIGSTOP);
1633 
1634 	/* Test that the modification is visible in ptrace after execution */
1635 	memset(xsave, 0xCC, xsave_size);
1636 	ret = ptrace(PTRACE_GETREGSET, child, (void *)NT_X86_XSTATE, &iov);
1637 	pkey_assert(ret == 0);
1638 	pkey_assert(*pkey_register == new_pkru);
1639 
1640 	/* Clear the PKRU bit from XSTATE_BV */
1641 	xstate_bv = (u64 *)(xsave + 512);
1642 	*xstate_bv &= ~(1 << 9);
1643 
1644 	ret = ptrace(PTRACE_SETREGSET, child, (void *)NT_X86_XSTATE, &iov);
1645 	pkey_assert(ret == 0);
1646 
1647 	/* Test that the modification is visible in ptrace before any execution */
1648 	memset(xsave, 0xCC, xsave_size);
1649 	ret = ptrace(PTRACE_GETREGSET, child, (void *)NT_X86_XSTATE, &iov);
1650 	pkey_assert(ret == 0);
1651 	pkey_assert(*pkey_register == 0);
1652 
1653 	ret = ptrace(PTRACE_CONT, child, 0, 0);
1654 	pkey_assert(ret == 0);
1655 
1656 	/* Test that the tracee saw the PKRU value go to 0 */
1657 	pkey_assert(child == waitpid(child, &status, 0));
1658 	dprintf3("[%d] waitpid(%d) status: %x\n", getpid(), child, status);
1659 	pkey_assert(WIFSTOPPED(status) && WSTOPSIG(status) == SIGSTOP);
1660 
1661 	/* Test that the modification is visible in ptrace after execution */
1662 	memset(xsave, 0xCC, xsave_size);
1663 	ret = ptrace(PTRACE_GETREGSET, child, (void *)NT_X86_XSTATE, &iov);
1664 	pkey_assert(ret == 0);
1665 	pkey_assert(*pkey_register == 0);
1666 
1667 	ret = ptrace(PTRACE_CONT, child, 0, 0);
1668 	pkey_assert(ret == 0);
1669 	pkey_assert(child == waitpid(child, &status, 0));
1670 	dprintf3("[%d] waitpid(%d) status: %x\n", getpid(), child, status);
1671 	pkey_assert(WIFEXITED(status));
1672 	pkey_assert(WEXITSTATUS(status) == 0);
1673 	free(xsave);
1674 }
1675 #endif
1676 
1677 void test_mprotect_pkey_on_unsupported_cpu(int *ptr, u16 pkey)
1678 {
1679 	int size = PAGE_SIZE;
1680 	int sret;
1681 
1682 	if (cpu_has_pkeys()) {
1683 		dprintf1("SKIP: %s: no CPU support\n", __func__);
1684 		return;
1685 	}
1686 
1687 	sret = syscall(SYS_mprotect_key, ptr, size, PROT_READ, pkey);
1688 	pkey_assert(sret < 0);
1689 }
1690 
1691 void (*pkey_tests[])(int *ptr, u16 pkey) = {
1692 	test_read_of_write_disabled_region,
1693 	test_read_of_access_disabled_region,
1694 	test_read_of_access_disabled_region_with_page_already_mapped,
1695 	test_write_of_write_disabled_region,
1696 	test_write_of_write_disabled_region_with_page_already_mapped,
1697 	test_write_of_access_disabled_region,
1698 	test_write_of_access_disabled_region_with_page_already_mapped,
1699 	test_kernel_write_of_access_disabled_region,
1700 	test_kernel_write_of_write_disabled_region,
1701 	test_kernel_gup_of_access_disabled_region,
1702 	test_kernel_gup_write_to_write_disabled_region,
1703 	test_executing_on_unreadable_memory,
1704 	test_implicit_mprotect_exec_only_memory,
1705 	test_mprotect_with_pkey_0,
1706 	test_ptrace_of_child,
1707 	test_pkey_init_state,
1708 	test_pkey_syscalls_on_non_allocated_pkey,
1709 	test_pkey_syscalls_bad_args,
1710 	test_pkey_alloc_exhaust,
1711 	test_pkey_alloc_free_attach_pkey0,
1712 #if defined(__i386__) || defined(__x86_64__)
1713 	test_ptrace_modifies_pkru,
1714 #endif
1715 };
1716 
1717 void run_tests_once(void)
1718 {
1719 	int *ptr;
1720 	int prot = PROT_READ|PROT_WRITE;
1721 
1722 	for (test_nr = 0; test_nr < ARRAY_SIZE(pkey_tests); test_nr++) {
1723 		int pkey;
1724 		int orig_pkey_faults = pkey_faults;
1725 
1726 		dprintf1("======================\n");
1727 		dprintf1("test %d preparing...\n", test_nr);
1728 
1729 		tracing_on();
1730 		pkey = alloc_random_pkey();
1731 		dprintf1("test %d starting with pkey: %d\n", test_nr, pkey);
1732 		ptr = malloc_pkey(PAGE_SIZE, prot, pkey);
1733 		dprintf1("test %d starting...\n", test_nr);
1734 		pkey_tests[test_nr](ptr, pkey);
1735 		dprintf1("freeing test memory: %p\n", ptr);
1736 		free_pkey_malloc(ptr);
1737 		sys_pkey_free(pkey);
1738 
1739 		dprintf1("pkey_faults: %d\n", pkey_faults);
1740 		dprintf1("orig_pkey_faults: %d\n", orig_pkey_faults);
1741 
1742 		tracing_off();
1743 		close_test_fds();
1744 
1745 		printf("test %2d PASSED (iteration %d)\n", test_nr, iteration_nr);
1746 		dprintf1("======================\n\n");
1747 	}
1748 	iteration_nr++;
1749 }
1750 
1751 void pkey_setup_shadow(void)
1752 {
1753 	shadow_pkey_reg = __read_pkey_reg();
1754 }
1755 
1756 int main(void)
1757 {
1758 	int nr_iterations = 22;
1759 	int pkeys_supported = is_pkeys_supported();
1760 
1761 	srand((unsigned int)time(NULL));
1762 
1763 	setup_handlers();
1764 
1765 	printf("has pkeys: %d\n", pkeys_supported);
1766 
1767 	if (!pkeys_supported) {
1768 		int size = PAGE_SIZE;
1769 		int *ptr;
1770 
1771 		printf("running PKEY tests for unsupported CPU/OS\n");
1772 
1773 		ptr  = mmap(NULL, size, PROT_NONE, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
1774 		assert(ptr != (void *)-1);
1775 		test_mprotect_pkey_on_unsupported_cpu(ptr, 1);
1776 		exit(0);
1777 	}
1778 
1779 	pkey_setup_shadow();
1780 	printf("startup pkey_reg: %016llx\n", read_pkey_reg());
1781 	setup_hugetlbfs();
1782 
1783 	while (nr_iterations-- > 0)
1784 		run_tests_once();
1785 
1786 	printf("done (all tests OK)\n");
1787 	return 0;
1788 }
1789